x86/GENERIC: Enable PCI_RESOURCE.
Has been needed for motherboards that are already several years old
by now, which require the OS to allocate PCI memory regions for some
devices like iwm(4); let's get some experience with having it on by
default.
PR port-amd64/59118: Thinkpad T495s - iwm PCI BAR is zero
make: in meta mode allow more general filtering
When a pre-built tree is cloned or relocated it is necessary to
filter all paths in .meta files that refer to the original location
to refer to the new.
For example (SB names the parent directory of src):
.if ${SB} != ${SB_ORIGIN:U${SB}}
.MAKE.META.FILTER += S,${SB_ORIGIN},${SB},
.endif
for commands; this is applied prior to any filters from
.MAKE.META.CMP_FILTER
but it is also applied to paths in the filemon syscall records
so we do not lose the benefit of them.
Reviewed by: rillig
pinentry: add a CHECK_PORTABILITY_SKIP
build-aux/config.rpath:286:12: double brackets
if [[ $libdir = '/usr/lib' ]] || [[ $libdir = '/usr/lib64' ]] ;
This file comes from gnulib and was perhaps generated with an
old version? gnulib master does not have the double brackets.
This code is under a case statement for GNU platforms so don't
bother patching it in pkgsrc.
atomic_store(9): Fix byte indexing on hash-locked architectures.
Fixes the easy part of:
PR kern/60393: hash-locked atomic_store_* is all kinds of busted
The hard part will require redoing these to just hash-lock the whole
store.
mk/compiler/gcc.mk: Require gcc 14 for c23
Note that NetBSD current has gcc 14. This is marked as temporary
because if a newer version is imported before branch, old current is
irrelevant.
While gcc documents that 15 is required for full support, we have a
long tradition of choosing a version that has almost full support,
when that results in fewer compiler builds. Guess that 14 will be ok,
with a comment indicating the choice is tentative.
Previously, putting c23 in USE_CC_FEATURES did nothing, so this commit
should not break any working packages. Tested on a random package not
to break, which is really about not having syntax errors.
mk/compiler/gcc.mk: Update C dialect comments
Be more precise about what gcc documents, including noting that C99
formally requires gcc 5 but we say 3 (and that it's been ok and use of
pre-5 is rare, so don't worry).
This is a comment-only change.
mk/compiler/gcc.mk: Update comments
- Note gcc12 in NetBSD 11.
- Explain that the point of limited versions is to build fewer
versions, not because is is simpler.
- Explain that long term we would prefer precise settings and a
per-platform limited list.
- Update gcc pointers, adding the C support matrix page and dropping
the C99 and C11 individual pages.
- Clarify that the gcc-current support page is about gcc-current,
not releases.
- Move C selection before C++.
- Sort C selection newest to oldest to align with C++ selection
sorting.
NFCI, and a single package test build showed no issues.
py-fido2: updated to 2.2.1
2.2.1 (released 2026-06-29)
* Server example: Migrate build tool from poetry to uv.
* Fix: Correctly format att_obj in previewSign.
Pass the actual mutex to mutex_spin_retry()
Avoid a situation where leftover splraise SR-mask cookie stays in a0.
Make UP kernel that double-acquires a spin mutex die with
"locking against myself" rather misleading garbage-pointer crash.
py-typing-extensions: updated to 4.16.0
4.16.0 (July 2, 2025)
No user-facing changes since 4.16.0rc2.
4.16.0rc2 (June 25, 2026)
- Avoid a `DeprecationWarning` when `deprecated` is applied to a coroutine function on
Python 3.14.0.
4.16.0rc1 (June 24, 2026)
- Make `typing_extensions.TypeAliasType`'s `__module__` attribute writable.
Backport of CPython PR
- Fix setting of `__required_keys__` and `__optional_keys__` when inheriting
keys with the same name.
- Add support for `AsyncIterator`, `io.Reader`, `io.Writer` and `os.PathLike` protocols
as bases for other protocols.
[28 lines not shown]
py-stevedore: updated to 5.9.0
5.9.0
* zuul: Use openstack-python3-next-jobs template
* pre-commit: Bump versions
* Do not install code to build release notes
* Drop support for Python 3.10
openvpn: updated to 2.7.5
2.7.5
mudp: send HMAC reset reply synchronously
options: fix use-after-free of DNS options on client connect
Ensure pushed tun-mtu is no lower than TUN_MTU_MIN
Make get_random return int64 instead of long
Clean up metadata handling in tls_crypt_v2_extract_client_key
Ensure we only get the session from valid tokens for external-auth
Ensure tls-crypt keys are not setup twice
dco_freebsd: Add check_malloc_return after realloc
Fix some msg() calls with wrong number of arguments in Windows-only code
openvpnserv: Address some uninitVariable warnings from cppcheck
openvpnserv: Fix memory leak when loading DLLs
renovate: Fix typo in regex manager
GHA: Maintenance update June 2026
GHA: Switch to using VS 2026
Fix 1-byte buffer overrun on NTLMv2 proxy responses.
[15 lines not shown]
